A System for Learning to Locomotion Using Adaptive Oscillators in the Humanoid Robot

This paper proposes a central pattern generators based control architecture using a frequency adaptive oscillator for learning to locomotion of humanoid robot. Central pattern generators are biological neural networks that can produce coordinated multidimensional rhythmic signals, under the control of simple input signals. They are found both in vertebrate and invertebrate animals for the control of locomotion. In this article, we present a novel system composed of adaptive nonlinear oscillators that can learn arbitrary rhythmic signals in a supervised learning framework, and apply it to control a simulated humanoid robot with up to 22 degrees of freedom. A key feature of the proposed architecture is that the learning is completely embedded in to the dynamical control, and does not require external optimization algorithms. As a test bed, we chose Robocup 3D soccer simulation environment (spark). Experimental results show that learn to walk of the robot could be successfully performed, thus allowing the biped robot to walk fast, stable and straightly.